Ok, after various issues (if your luck is as legendarily bad as mine, don't start building a computer on a friday the 13th) I now have my Bulldozer rig working in a way I'm fairly happy with.
For the time being, the graphics card is an HD4830 that I had lying around, till the HD7770 comes along (and my bank account is replenished). By extrapolating the results of Micheal benchmarks with fglrx I'd say it's equivalent to the HD6670 performance-wise.
Right now dual-booting Xubuntu 11.10 32 bit and openSUSE 12.1 64 bit.
Quick observations so far:
- Turbo Core isn't working properly in any of those distros, although I've enabled it in UEFI. Frequency never goes over the nominal 3.3GHz on any of the cores.
- Temperature monitoring is a bit all over the place (with 10-15C spikes when loading a program), though I believe that it's because idle temps are wrong. Even with standard cooler and all cores maxed, the highest I've seen is 50C though. Which I find quite decent.
- Quick Unigine tests show a noticeable difference in performance between 32 and 64 bit: 950 vs 1130 respectively in windowed mode with tesselation off. The lower 32 bit score is the same in both Xubuntu and openSUSE. GCC-32bit lacking optimizations?
Of course I'll run PTS as soon as I get the time, though I thought you guys might want something specific?
Looks like yes. Please look into BIOS and kernel develop guide for AMD Family 12h processors, "126.96.36.199.1 Core Performance Boost (CPB)" section. Real frequency is visible in cpufreq-aperf and x86info. You can check enabled CPB or not:Code:for i in `find /sys/ -name cpb`; do echo -n "$i: "; cat $i; done
for i in `find /sys/ -name cpb`; do echo -n "$i: "; cat $i; done
I've got a Phenom II x6 1055t as well, but I'm not at home and the computer is currently taking a nap.
FWIW, my Phenom X6 says all six cores have cbp enabled, under kernel 3.0.9.
The main issue I believe isn't whether CBP is enabled or not, but rather whether it's working like it should or not.
Turbo Core level is a function of the power draw and the number of cores under load, correct?
Michael's test with the FX-8150 show improvements on some single threaded benchmarks, but the information feels overall rather incomplete, partly due to the lack of proper monitoring.
i got fx6100 ubu10.10
i turned off turbo boost and use cpu frequency scaling monitor 2.30.0
the latest amd chips have on chip power and thermal comtrols
my voltages jump up and down
turbo mode turns off cores so if your monitoring one core you may not see it
i think its better to not defeat these on chip controls